Re: WARNING: CPU: 1 PID: 83 at arch/x86/kernel/cpu/sgx/main.c:446 ksgxd+0x1b7/0x1d0

[Jarkko Sakkinen <jarkko@kernel.org>]

On Wed, Aug 24, 2022 at 09:12:06PM -0500, Haitao Huang wrote:
> Hi Paul
> 
> On Tue, 23 Aug 2022 08:48:52 -0500, Paul Menzel <pmenzel@molgen.mpg.de>
> wrote:
> 
> > Dear Dave,
> > 
> > 
> > Am 20.08.22 um 08:13 schrieb Paul Menzel:
> > 
> > > Am 19.08.22 um 20:28 schrieb Dave Hansen:
> > > > On 8/19/22 09:02, Paul Menzel wrote:
> > > > > On the Dell XPS 13 9370, Linux 5.18.16 prints the warning below:
> > > > > 
> > > > > ```
> > > > > [    0.000000] Linux version 5.18.0-4-amd64
> > > > > (debian-kernel@lists.debian.org) (gcc-11 (Debian 11.3.0-5)
> > > > > 11.3.0, GNU ld (GNU Binutils for Debian) 2.38.90.20220713)
> > > > > #1 SMP PREEMPT_DYNAMIC Debian 5.18.16-1 (2022-08-10)
> > > > > [    0.000000] Command line:
> > > > > BOOT_IMAGE=/vmlinuz-5.18.0-4-amd64
> > > > > root=UUID=56f398e0-1e25-4fda-aa9f-611dece4b333 ro quiet
> > > > > […]
> > > > > [    0.000000] DMI: Dell Inc. XPS 13 9370/0RMYH9, BIOS
> > > > > 1.21.0 07/06/2022
> > > > > […]
> > > > > [    0.235418] sgx: EPC section 0x40200000-0x45f7ffff
> > > 
> > > > Would you be able to send the entire dmesg, along with:
> > >  The log message are attached to the first message, where I missed
> > > to carbon-copy linux-sgx@ [1].
> > > 
> > > >     cat /proc/iomem # (as root)
> > > > and
> > > >     cpuid -1 --raw
> > >  I am going to provide that next week. (Side note, Intel might have
> > > some Dell XPS 9370 test machines in some QA lab.)
> > 
> > Please find both outputs at the end of the file.
> > 
> 
> Could you also check output of "sudo rdmsr -x 0x3a"?
> Also was CONFIG_X86_SGX_KVM set?
> 
> If CONFIG_X86_SGX_KVM is not set and bit 17 (SGX_LC) of the MSR 3A not set,
> then I think following sequence during sgx_init is possible:
> 
> sgx_page_cache_init -> sgx_setup_epc_section
>                        ->put all physical EPC pages in sgx_dirty_page_list.
> Kick off ksgxd.
> Later, sgx_drv_init returns none-zero due to this check:
>     if (!cpu_feature_enabled(X86_FEATURE_SGX_LC))
>         return -ENODEV;
> sgx_vepc_init also returns none-zero if CONFIG_X86_SGX_KVM was not set.
> 
> And sgx_init will call kthread_stop(ksgxd_tsk):
>     ret = sgx_drv_init();
> 
>     if (sgx_vepc_init() && ret)
>         goto err_provision;
> ...
> err_provision:
>     misc_deregister(&sgx_dev_provision);
> 
> err_kthread:
>     kthread_stop(ksgxd_tsk);
> 
> 
> That triggers __sgx_sanitize_pages return early due to these lines:
>     /* dirty_page_list is thread-local, no need for a lock: */
>     while (!list_empty(dirty_page_list)) {
>         if (kthread_should_stop())
>             return;
> 
> And that would trigger (depends on timing?) the warning in ksgxd due to
> non-empty sgx_dirty_page_list
> at that moment.

You're correct, and it's not a bug but completely legit behaviour.

And given that non-empty dirty page list is legit behavior WARN_ON()
is not what should be used in here.

Fix coming in a bit.

BR, Jarkko